replay-ux-research

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted telemetry data from external users.
  • Ingestion points: The agent retrieves data through search_events and get_replay_details tools (in SKILL.md), which include user-controlled query parameters, URLs, and activity breadcrumbs from real session replays.
  • Boundary markers: The skill lacks explicit delimiters or instructions to treat data retrieved from the Sentry API as untrusted, increasing the risk that the agent may follow instructions embedded within user-provided strings (e.g., in a URL search parameter).
  • Capability inventory: The skill's environment includes tools for querying Sentry events and resources (search_events, get_replay_details, search_issues, get_sentry_resource) but does not grant access to sensitive local files, shell command execution, or arbitrary network requests beyond the Sentry environment.
  • Sanitization: While the instructions mandate anonymizing user email addresses, there is no validation or filtering applied to other user-generated content like navigation paths or search queries before they are incorporated into the final report.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 01:44 PM
Security Audit — agent-trust-hub — replay-ux-research